Thanks to Backtick, we were able to incorporate AI into our product with ease. Their flexibility in adapting to our startup’s time and budget constraints, combined with their expertise in AI and ability to integrate their solution into our existing infrastructure, made for a smooth and valuable experience.

Jennie Orton

CEO, bintel

INTRODUCTION

Bintel is a company providing predictive, data driven waste management services. Their motto is simple - no time to waste. We’re proud of being a part of Bintels journey to create a more sustainable future, ensuring we’re doing everything possible to reduce unnecessary emissions by making clever use of next generation technologies.

CHALLENGE

Binel provides and install IoT devices on waste containers. This allows collection of live fill level data. The data is a core component in next generation waste management where entire systems are monitored and managed on a on-demand basis as opposed to static collection schedules used today. For example, a truck is sent to empty waste bins in a specific area every Thursday.

GOAL

Our task was to ultimately build routing algorithms for a fleet of trucks to optimize waste collection schedules. Trucks have different capacities - not just in volume, but different capabilities with regards to what fractions the can empty (plastic, metal etc).

SOLUTION

Backtick collaborates with bintel on multiple projects. We’re the studio responsible for bintel’s ML infrastructure and models. The models are continuously predicting future waste container fill levels and detecting anomalies in sensor data, ensuring high quality data streams to downstream services.

In order to solve the challenge of smarter waste collection schedules, we modeled and developed a few building blocks that integrated with bintel's data streams and infrastructure:

• Anomaly detection for incoming data - while IoT devices are great, sometimes reality messes things up - like when people throw trash with compressing it, or a spider web that messes with the measurements

• Machine learning models to predict when waste containers are full

• Vehicle routing for a fleet of trucks with regards to capacity, capabilities, the predicted fill levels and certain time windows regarding truck available.

RESULTS

Multiple machine learning models trained and managed with MLFlow and deployed in Docker containers at a Swedish cloud provider. Implementation of the capacitated vehicle routing problem with time windows (CVRPTW).